November 19, 2026

Beyond the Answering Machine

AI is extremely easy to talk about. Extremely hard to implement. There's so much hype around it that the distance between those two things gets lost. Then you sit down and try to make yourself genuinely productive with it, and the mess becomes obvious.

First, without tools, AI isn't going to get much done for you. Then you give it tools and discover the other limiting factor. You.

I think this is where the conversation needs to move beyond individual productivity. An engineer showing how much faster they can build is useful. Don't get me wrong, I love that stuff. But AI reaching its potential in pockets doesn't mean a company has changed. For that, it has to become part of how the company works, culturally as well as operationally.

And that starts with something painfully unglamorous. Defining your universe. What tools does the company use? Where does information live? Who has access to what? What can an agent safely read, and what can it safely do? You have to figure out access, permissions, and security before you can hand those tools to company-level agents. Not a demo setup. Something people can actually use safely.

Get all of that right and you can do some great knowledge work. Ask deep questions, connect information across tools, get answers that would have taken someone hours to piece together. That's useful. Very useful. But you may still have built a great answering machine.

Getting shit done is another hurdle.

Now you have to painfully encode how your company actually works. How your systems behave. How your repositories are organized. How deployments happen. How customer workflows run. The tools might all be connected to the agent, but how do they connect to each other inside a workflow? Having access to every tool involved in a process is not the same as knowing how to carry out that process.

This is the work that's hard to show in a productivity demo. Making the processes explicit, working through the details, giving the agent enough context to act rather than just explain. It's painful enough that I think it takes a real jolt for a company to invest in it meaningfully. Not just give people access to AI and hope the rest sorts itself out.

Cross that hurdle and, to me, that's nirvana. Not because the model suddenly got smarter, but because you've finally done the work that lets it be useful across the company.

Most of the narrative I see is still an engineer or a builder showing how productive they've become. That matters. But until we unlock this at a team or company level, I don't think we're seeing the true potential yet. We're seeing what motivated individuals can make it do. The harder question is what it takes for everyone to work differently.